Did Brookings Get It Wrong?

In the second half of 2020, American society became considerably more violent.  For example, in New York City there were 187 murders in the first six months of 2020, and 276 in the second half. (For more detailed city-by-city and month-by-month data, I suggest the American Violence website). During the latter period, homicides rose in every American city with over 500,000 people for which data is available, and rose by over 10 percent in all but one of these cities.  (In the lone exception, Baltimore, homicides rose by 7 percent).  My initial suspicion was that this violence was somehow related to anti-police agitation that led to and followed the George Floyd protests of May and June 2020- perhaps because police were less aggressive, or because civilians were less willing to cooperate with police. 

However, the Brookings Institution has come out with a paper that blames the American crime wave on the aftereffects of COVID.   The Brookings paper asserts that homicides began to rise in mid-April as COVID restrictions eased, and continued to rise throughout 2020.  But rather than blaming government criminal justice policies or anti-police hostility, Brookings blames unemployment and school closures that left young men on the streets, claiming that the spike in murders was “directly connected to local unemployment and school closures.”  This conclusion is comfort food for both the Left and Right: on the Left, because leftists tend to blame crime on unemployment and inadequate social services, and on the Right, because conservatives have generally complained more aggressively than liberals about COVID-era business and school closures.

But it seems to me that there are three problems with this analysis.  First of all, historically there has not always been a strong correlation in the U.S. between unemployment and crime. The Brookings paper linked to data showing that unemployment among high school dropouts rose from under 6 percent in Feb. 2020 to over 20 percent in April, gradually returning to its prior level in early 2022.  However, the same data shows that during the 2006-2010 recession, unemployment among this group rose from under 6 percent to about 15 percent.  Yet the American homicide rate actually declined during those years, from 5.7 per 100,000 Americans to 4.8. (By contrast, in 2020-22 the homicide rate was between 6 and 7 homicides per 100,000 Americans).   So it seems clear that unemployment alone was not an important factor in 2020.

The Brookings paper solves this problem by tying unemployment and school closures together, because of the combination of "young men forced out of work and teen boys pushed out of school in low-income neighborhoods".  If the latter was highly relevant, homicides by Americans under 20 would have increased more rapidly than homicides by grownups, because the latter group was left idle by school closures.  The FBI "Homicide Explorer" website allows us to differentiate homicides by offender age.  Brookings asserts that the rise in homicides began in April, so to test this theory I compared April-July homicides in 2019 to April-July homicides in 2020.  During the first period, there were 757 homicide offenders under 20 and 3413 over 20 (if I am counting correctly).  During the later period, there were 967 under-20 offenders and 4658 over-20 offenders.  In other words, between April-June 2019 and April-June 2020 there was a 36 percent increase in over-20 offenders and a 27 percent increase among over-20 offenders.  Thus, homicide actually rose less rapidly among under-20 Americans, which suggests that school closures and other factors peculiar to juveniles were not relevant during these months. 

Second, Brookings emphasizes that homicide rates were already growing before George Floyd’s death in late May.  There is an element of truth to this: the FBI Homicide Explorer shows that homicide grew by almost 20 percent (from 1129 to 1334) between April 2019 and April 2020, and by over 20 percent (from 1314 to 1583) between May 2019 and May 2020.  

But what happened after the Floyd riots?  In June, murders increased from 1303 in 2019 to 1811 in 2020- almost a 40 percent increase.  In July, they increased from 1342 in 2019 to 1895 in 2020- over a 40 percent increase.  In August, they increased from 1290 in 2019 to 1820 in 2020- over a 40 percent increase.  In September, October, November, and December 2020, the number of murders increased by between 420 and 528 compared to comparable months in 2019.  In other words, if homicide rates had continued to grow at an April 2020 level, there would have been about 1400 “extra” murders between June and December 2020 (compared to their 2019 level).  Instead, there were almost 3500. 

I double-checked this data by looking at the American Violence webpage, which has city-by-city data.  I began by comparing April-May 2019 to April-May 2020.    Among cities with over 500,000 population, homicides increased in eighteen of the twenty-eight cities for which data was available; in four of the twenty-eight, the number of homicides increased by only one (e.g. from six to seven) which does not seem particularly significant.   In other words, the median increase (i.e. the level at which half were above and half were below) was one homicide.  Then I compared June-July 2019 to June-July 2020.  Murders increased in twenty-five of the twenty-eight cities, and increased by more than one in twenty-two of them.   

More startling to me was the level of increase in homicides after the riots.  Murders doubled or more in only two cities between April-May 2019 and April-May 2020. Murders doubled or more in seven cities in June/July.  If I am reading the data correctly the median April-May increase was 16 percent, while the median June-July homicide increase was 50 percent.  In sum, it seems very clear that while homicides did rise in large cities in March and April, they rose much more (compared to 2019) in June and July after the protests and riots.  So if the Brookings paper is trying to argue that the riots changed nothing, that argument is not supported by the data.

Third, other democracies experienced high unemployment rates and school closures, and the Brookings study admits that those democracies did not experience crime waves.  But the Brookings answer is “but … guns.”  Of course, the U.S. was awash in guns long before COVID, so it may seem hard to believe that guns are responsible for a quick, sharp, increase in homicide of the sort that occurred in 2020.  But the Brookings paper tries to make this argument, by asserting that “Because the U.S. spike in homicides was due to gun violence, it is not surprising to see that countries that do not struggle with this problem avoided similar increases in murder—particularly during a time when being close enough to stab or physically assault someone carried the chance of being exposed to a deadly virus.”  In other words, the Brookings argument seems to be: if Americans didn’t have guns, they would not have committed murders because they were afraid to stab each other.   This counterfactual seems impossible to disprove. 

Having said that, I decided to take a look at the data to see if they say anything relevant. If I am reading FBI data on crimes by weapon correctly, handgun murders increased from 2331 in April-July 2019 to 3063 in April-July 2020, a 31 percent increase. If you add the “other firearms” category (which means, I suspect, that police were unsure what type of firearm was involved, as opposed to firearms clearly involving long guns and shotguns*) the number of “cheap firearm” homicides increased from 3475 to 4749, a 36 percent increase.  Knife homicides went up from 515 to 642, a 24 percent increase.  So gun murders did go up by more than knife murders - but the latter still grew at an alarming rate.  During the two months after the George Floyd protests, knife homicides increased from 270 to 377, almost a 40 percent increase. 

If the COVID-era crime problem was the result of a nation awash in guns, I would have thought that the most gun-happy parts of the nation would experience the largest homicide increases.  Homicide data from American Violence suggests that there are at least a few parts of the nation where a significant number of murders do not involve guns.   Legal gun ownership is rare in New York City, and the American Violence site suggests that the number of fatal shootings there was only about 60 percent of overall number of homicides in mid-2020 (as opposed to about 80 percent in cities like Chicago and Philadelphia).  However, homicides rose as rapidly in New York as they did in other cities.  Having said that, I am reluctant to generalize from a sample of one, and more research on gun ownership by city might be useful. 

In sum, it seems unlikely to me that unemployment and school closures were a cause of the 2020 crime wave.  It also seems clear to me (as the Brookings paper suggests) that homicides increased in the months before the protests - but also that they increased even more in the months after the protests.  It also appears that homicides were somewhat more likely to involve guns in 2020 than in 2019- but it also appears that even knife homicides increased. 

*I note that despite the public controversy about so-called "assault rifles" these guns are a pretty small part of the U.S. crime problem; only 265 homicides during March-July 2020 involved rifles or shotguns.